That one uses a collapsed gibbs sampler, ours is variational. We should change the reference to be explicit about that. ping @ariddell

I didn't know there was a reference. Thanks!

The algorithms (Gibbs vs. VB) are very different.

Do you want to sent a PR that explains the difference in one or two sentences? It is more for the interested expert, I guess.
Otherwise we'll find someone to do it ;)

Sure. I'll write a sentence or two. I need to add a link to the sklearn
implementation anyway (in the lda repo) for those who want to use VB.

On 10/22, Andreas Mueller wrote:

Do you want to sent a PR that explains the difference in one or two sentences? It is more for the interested expert, I guess.
Otherwise we'll find someone to do it ;)

---

Reply to this email directly or view it on GitHub:
#5529 (comment)